From owner-freebsd-current  Wed Aug  6 04:58:22 1997
Return-Path: <owner-freebsd-current>
Received: (from root@localhost)
          by hub.freebsd.org (8.8.5/8.8.5) id EAA18879
          for current-outgoing; Wed, 6 Aug 1997 04:58:22 -0700 (PDT)
Received: from broon.off.connect.com.au (broon.off.connect.com.au [203.63.69.1])
          by hub.freebsd.org (8.8.5/8.8.5) with ESMTP id EAA18874
          for <current@freebsd.org>; Wed, 6 Aug 1997 04:58:10 -0700 (PDT)
Received: from connect.com.au (ggm@localhost) by broon.off.connect.com.au with ESMTP id VAA23819
  (8.8.5/IDA-1.6); Wed, 6 Aug 1997 21:57:02 +1000 (EST)
To: Bruce Evans <bde@zeta.org.au>
cc: msmith@atrad.adelaide.edu.au, current@freebsd.org
Subject: Re: canonical issue wrt DMA & wd/wdc 
In-reply-to: Your message of "Wed, 06 Aug 1997 21:30:42 +1000."
             <199708061130.VAA15302@godzilla.zeta.org.au> 
Date: Wed, 06 Aug 1997 21:57:00 +1000
Message-ID: <23816.870868620@connect.com.au>
From: George Michaelson <ggm@connect.com.au>
Sender: owner-freebsd-current@freebsd.org
X-Loop: FreeBSD.org
Precedence: bulk


I applied them, made a kernel and ran bonnie single user under each
on a fujitsu 1.2Gb drive. here is the outcomes:

              -------Sequential Output-------- ---Sequential Input-- --Random--
              -Per Char- --Block--- -Rewrite-- -Per Char- --Block--- --Seeks---
Machine    MB K/sec %CPU K/sec %CPU K/sec %CPU K/sec %CPU K/sec %CPU  /sec %CPU
p100-no3   20  2285 63.1  2715 11.0  2528 19.1  3820 96.5 36465 100.1 825.4 22.9
p100-no3   20  2340 65.6  2748 11.1  2459 17.5  3830 96.3 36437 100.0 860.8 23.0
p100-no3   20  2378 63.4  4863 18.8  5076 29.2  3956 99.1 36521 100.1 781.8 21.4
p100-no3   20  2462 65.3  4666 18.1  5384 33.6  3949 99.1 36485 100.1 788.5 21.6

p100+32    20  2879 77.9  6762 24.7  6655 39.8  3842 96.4 29165 100.0 798.9 22.6
p100+32    20  2900 78.9  5961 25.5  7216 42.6  3969 99.2 29109 98.9 789.9 23.0
p100+32    20  2821 77.7  4594 19.2  4592 29.8  3838 95.8 28799 99.9 809.3 23.7
p100+32    20  2821 78.2  4589 20.8  4607 27.2  3828 96.1 28850 100.0 784.8 23.8

no3 is no flags. +32 is with flags 80ff.

Interesting that while per-char and sequential block output had slight
improvement, input gets marginally worse for flagged disk!

I'd say that on this evidence, the 32-bit and multi-sector transfers are
of marginal benefit. The above doesn't really show clear wins. Does it?

Asus MB 512k with P100, 32Mb non-EDO ram. 

-George